Is it normal that saying "hello" isn't detected well?

[qwbarch]

I noticed that when I say "hello" the probability is always very low. If I say "hello, test test", the part where i say "test test" is completely fine, whereas the "hello" has a low probability as usual.

This is on both the v4 and v5 model. Is this expected behavior?

[snakers4]

Can you send an audio example?

[qwbarch]

Whoops, sorry for not providing an example. Here is an audio clip:
https://drive.google.com/file/d/1MvX4tMAhDjtzq3mI32JJgt2r5-VzPiaP/view?usp=sharing

On the v4 model, this will still have a low probability (around 0.1f-0.3f if I remember right), but on the v5 model it seems to think there's no speech detected at all

[snakers4]

Loading different versions like this:

  model, utils = torch.hub.load(repo_or_dir='snakers4/silero-vad:v3.1',
                                model='silero_vad',
                                force_reload=True,
                                onnx=USE_ONNX)

  (get_speech_timestamps,
  save_audio,
  read_audio,
  VADIterator,
  collect_chunks) = utils

v3.1

v4.0

v5.1

So, in your particular case, I would suggest these params:

speech_timestamps = get_speech_timestamps(wav, model, sampling_rate=SAMPLING_RATE, visualize_probs=True, threshold=0.15)